arXiv subjects

Wenkai Yang

Publications and source records attributed to Wenkai Yang.

At least 37 records · Page 2Linked to original sources

arXiv2025

LaSeR: Reinforcement Learning with Last-Token Self-Rewarding

Reinforcement Learning with Verifiable Rewards (RLVR) has recently emerged as a core paradigm for enhancing the reasoning capabilities of Large Language Models (LLMs). To address the lack of verification signals at test time, prior studies incorporate the training of model's self-verification capability into the standard RLVR process, thereby unifying reasoning and verification capabilities within a single LLM. However, previous practice requires the LLM to sequentially generate solutions and self-verifications using two separate prompt templates, which significantly reduces efficiency. In this work, we theoretically reveal that the closed-form solution to the RL objective of self-verification can be reduced to a remarkably simple form: the true reasoning reward of a solution is equal to its last-token self-rewarding score, which is computed as the difference between the policy model's next-token log-probability assigned to any pre-specified token at the solution's last token and a pre-calculated constant, scaled by the KL coefficient. Based on this insight, we propose LaSeR (Reinforcement Learning with Last-Token Self-Rewarding), an algorithm that simply augments the original RLVR loss with a MSE loss that aligns the last-token self-rewarding scores with verifier-based reasoning rewards, jointly optimizing the reasoning and self-rewarding capabilities of LLMs. The optimized self-rewarding scores can be utilized in both training and testing to enhance model performance. Notably, our algorithm derives these scores from the predicted next-token probability distribution of the last token immediately after generation, incurring only the minimal extra cost of one additional token inference. Experiments show that our method not only improves the model's reasoning performance but also equips it with remarkable self-rewarding capability, thereby boosting its inference-time scaling performance.

cs.CL↗

arXiv2025

Towards Thinking-Optimal Scaling of Test-Time Compute for LLM Reasoning

Recent studies have shown that making a model spend more time thinking through longer Chain of Thoughts (CoTs) enables it to gain significant improvements in complex reasoning tasks. While current researches continue to explore the benefits of increasing test-time compute by extending the CoT lengths of Large Language Models (LLMs), we are concerned about a potential issue hidden behind the current pursuit of test-time scaling: Would excessively scaling the CoT length actually bring adverse effects to a model's reasoning performance? Our explorations on mathematical reasoning tasks reveal an unexpected finding that scaling with longer CoTs can indeed impair the reasoning performance of LLMs in certain domains. Moreover, we discover that there exists an optimal scaled length distribution that differs across different domains. Based on these insights, we propose a Thinking-Optimal Scaling strategy. Our method first uses a small set of seed data with varying response length distributions to teach the model to adopt different reasoning efforts for deep thinking. Then, the model selects its shortest correct response under different reasoning efforts on additional problems for self-improvement. Our self-improved models built upon Qwen2.5-32B-Instruct outperform other distillation-based 32B o1-like models across various math benchmarks, and achieve performance on par with the teacher model QwQ-32B-Preview that produces the seed data.

cs.CL↗

arXiv2025

Enhanced photon-pair generation from a van der Waals metasurface

Quantum photon pairs play a pivotal role in many quantum applications. Metasurfaces, two-dimensional arrays of nanostructures, have been studied intensively to enhance and control pair generation via spontaneous parametric downconversion (SPDC). Van der Waals (VdW) layered materials have emerged as promising candidates for nonlinear materials in quantum light sources, owing to their high nonlinear susceptibility and compatibility with on-chip integration. In this work, we present the first demonstration of SPDC from a metasurface composed of the VdW material 3R-MoS2. The nanoresonators support quasi-bound states in the continuum (qBIC) with a quality factor of up to 120, enhancing light-matter interactions. This design achieves a 20-fold increase in SPDC rate compared to an unstructured film and significantly higher brightness, resulting in enhanced quantum photon-pair generation. This work establishes a new approach for utilizing van der Waals metasurfaces in the generation of quantum photon pairs, opening avenues for advanced quantum applications.

physics.optics↗

arXiv2025

The Capabilities and Limitations of Weak-to-Strong Generalization: Generalization and Calibration

Weak-to-strong generalization, where weakly supervised strong models outperform their weaker teachers, offers a promising approach to aligning superhuman models with human values. To deepen the understanding of this approach, we provide theoretical insights into its capabilities and limitations. First, in the classification setting, we establish upper and lower generalization error bounds for the strong model, identifying the primary limitations as stemming from the weak model's generalization error and the optimization objective itself. Additionally, we derive lower and upper bounds on the calibration error of the strong model. These theoretical bounds reveal two critical insights: (1) the weak model should demonstrate strong generalization performance and maintain well-calibrated predictions, and (2) the strong model's training process must strike a careful balance, as excessive optimization could undermine its generalization capability by over-relying on the weak supervision signals. Finally, in the regression setting, we extend the work of Charikar et al. (2024) to a loss function based on Kullback-Leibler (KL) divergence, offering guarantees that the strong student can outperform its weak teacher by at least the magnitude of their disagreement. We conduct sufficient experiments to validate our theory.

cs.LG↗

arXiv2025

On Weak-to-Strong Generalization and f-Divergence

Weak-to-strong generalization (W2SG) has emerged as a promising paradigm for stimulating the capabilities of strong pre-trained models by leveraging supervision from weaker supervisors. To improve the performance of the strong model, existing methods often require additional weak models or complex procedures, leading to substantial computational and memory overhead. Motivated by the effectiveness of $f$-divergence loss in various machine learning domains, we introduce $f$-divergence as an information-theoretic loss function framework in W2SG. Our theoretical analysis reveals fundamental limitations and equivalence of different $f$-divergence losses in W2SG, supported by sample complexity bounds and information-theoretic insights. We empirically demonstrate that $f$-divergence loss, which generalizes widely-used metrics like KL divergence, effectively improves generalization and noise tolerance of the strong model in practice.

cs.LG↗

arXiv2025

Revisiting Weak-to-Strong Generalization in Theory and Practice: Reverse KL vs. Forward KL

As large language models advance toward superhuman performance, ensuring their alignment with human values and abilities grows increasingly complex. Weak-to-strong generalization offers a promising approach by leveraging predictions from weaker models to guide stronger systems, but its effectiveness could be constrained by the inherent noise and inaccuracies in these weak predictions. To address this, we propose a theoretically grounded approach that replaces forward KL divergence-whose mass-covering behavior risks overfitting to imperfect weak signals-with reverse KL divergence. Reverse KL divergence's zero-forcing effect prioritizes high-confidence predictions, effectively mitigating the influence of unreliable weak supervision. Theoretically, we extend existing bounds and derive tighter lower bounds for both forward and reverse KL divergence, establishing that reverse KL achieves at least comparable guarantees to forward KL. Notably, when a sufficiently pre-trained strong model is fine-tuned on the last linear layer, reverse KL guarantees that it outperforms its weak supervisor by the magnitude of their disagreement. Empirically, we demonstrate that reverse KL and reverse cross-entropy enable strong models to successfully outperform those trained with forward KL and standard cross-entropy across most settings, highlighting the practical advantages of these reverse losses.

cs.LG↗

arXiv2025

DeepCritic: Deliberate Critique with Large Language Models

As Large Language Models (LLMs) are rapidly evolving, providing accurate feedback and scalable oversight on their outputs becomes an urgent and critical problem. Leveraging LLMs as critique models to achieve automated supervision is a promising solution. In this work, we focus on studying and enhancing the math critique ability of LLMs. Current LLM critics provide critiques that are too shallow and superficial on each step, leading to low judgment accuracy and struggling to offer sufficient feedback for the LLM generator to correct mistakes. To tackle this issue, we propose a novel and effective two-stage framework to develop LLM critics that are capable of deliberately critiquing on each reasoning step of math solutions. In the first stage, we utilize Qwen2.5-72B-Instruct to generate 4.5K long-form critiques as seed data for supervised fine-tuning. Each seed critique consists of deliberate step-wise critiques that includes multi-perspective verifications as well as in-depth critiques of initial critiques for each reasoning step. Then, we perform reinforcement learning on the fine-tuned model with either existing human-labeled data from PRM800K or our automatically annotated data obtained via Monte Carlo sampling-based correctness estimation, to further incentivize its critique ability. Our developed critique model built on Qwen2.5-7B-Instruct not only significantly outperforms existing LLM critics (including the same-sized DeepSeek-R1-distill models and GPT-4o) on various error identification benchmarks, but also more effectively helps the LLM generator refine erroneous steps through more detailed feedback.

cs.CL↗

arXiv2025

Giant Second Harmonic Generation from 3R-MoS$_2$ Metasurfaces

Metasurfaces have long served as a cornerstone technique to enhance nonlinear processes, enabling frequency conversion, efficient light manipulation and integrated photonic devices. However, traditional bulk materials often suffer from high absorption losses, hindering the second harmonic generation (SHG) efficiency. Here, we develop a novel approach exploiting quasi-bound state in continuum (qBIC) to achieve giant SHG efficiency in metasurfaces utilizing 3R-MoS$_2$, with high index, superior damage threshold and inherent nonlinearity. The high refractive index of 3R-MoS$_2$, facilitates the high-quality factor (Q) metasurfaces, leading to reduced radiation leakage and localized light confinement within qBIC resonances, with which a remarkable 2000-fold enhancement in SHG intensity has been experimentally demonstrated. Additionally, the twist angle between the lattice orientation and the metasurface unit geometry exhibits a 120$^\circ$ periodicity in its influence on SHG behaviour. By strategically designing to realize the qBIC and exciton dual resonances and optimized twist angle (30$^\circ$), SHG conversion efficiency was boosted to ~1%, which is around 2 orders of magnitude higher than those of the best metasurfaces on traditional bulk materials. This approach enables potential applications in various areas of nonlinear optics, including frequency conversion, light manipulation, integrated photonics, and quantum communications.

physics.app-ph↗

arXiv2025

Super(ficial)-alignment: Strong Models May Deceive Weak Models in Weak-to-Strong Generalization

Superalignment, where humans act as weak supervisors for superhuman models, has become a crucial problem with the rapid development of Large Language Models (LLMs). Recent work has preliminarily studied this problem by using weak models to supervise strong models, and discovered that weakly supervised strong students can consistently outperform weak teachers towards the alignment target, leading to a weak-to-strong generalization phenomenon. However, we are concerned that behind such a promising phenomenon, whether there exists an issue of weak-to-strong deception, where strong models deceive weak models by exhibiting well-aligned in areas known to weak models but producing misaligned behaviors in cases weak models do not know. We take an initial step towards exploring this security issue in a specific but realistic multi-objective alignment case, where there may be some alignment targets conflicting with each other (e.g., helpfulness v.s. harmlessness). We aim to explore whether, in such cases, strong models might deliberately make mistakes in areas known to them but unknown to weak models within one alignment dimension, in exchange for a higher reward in another dimension. Through extensive experiments in both the reward modeling and preference optimization scenarios, we find: (1) The weak-to-strong deception phenomenon exists across all settings. (2) The deception intensifies as the capability gap between weak and strong models increases. (3) Bootstrapping with an intermediate model can mitigate the deception to some extent, though its effectiveness remains limited. Our work highlights the urgent need to pay more attention to the true reliability of superalignment.

cs.CL↗

arXiv2024

Distilling Rule-based Knowledge into Large Language Models

Large language models (LLMs) have shown incredible performance in completing various real-world tasks. The current paradigm of knowledge learning for LLMs is mainly based on learning from examples, in which LLMs learn the internal rule implicitly from a certain number of supervised examples. However, this learning paradigm may not well learn those complicated rules, especially when the training examples are limited. We are inspired that humans can learn the new tasks or knowledge in another way by learning from rules. That is, humans can learn new tasks or grasp new knowledge quickly and generalize well given only a detailed rule and a few optional examples. Therefore, in this paper, we aim to explore the feasibility of this new learning paradigm, which targets on encoding rule-based knowledge into LLMs. We further propose rule distillation, which first uses the strong in-context abilities of LLMs to extract the knowledge from the textual rules, and then explicitly encode the knowledge into the parameters of LLMs by learning from the above in-context signals produced inside the model. Our experiments show that making LLMs learn from rules by our method is much more efficient than example-based learning in both the sample size and generalization ability. Warning: This paper may contain examples with offensive content.

cs.CL↗

arXiv2024

Watch Out for Your Agents! Investigating Backdoor Threats to LLM-Based Agents

Driven by the rapid development of Large Language Models (LLMs), LLM-based agents have been developed to handle various real-world applications, including finance, healthcare, and shopping, etc. It is crucial to ensure the reliability and security of LLM-based agents during applications. However, the safety issues of LLM-based agents are currently under-explored. In this work, we take the first step to investigate one of the typical safety threats, backdoor attack, to LLM-based agents. We first formulate a general framework of agent backdoor attacks, then we present a thorough analysis of different forms of agent backdoor attacks. Specifically, compared with traditional backdoor attacks on LLMs that are only able to manipulate the user inputs and model outputs, agent backdoor attacks exhibit more diverse and covert forms: (1) From the perspective of the final attacking outcomes, the agent backdoor attacker can not only choose to manipulate the final output distribution, but also introduce the malicious behavior in an intermediate reasoning step only, while keeping the final output correct. (2) Furthermore, the former category can be divided into two subcategories based on trigger locations, in which the backdoor trigger can either be hidden in the user query or appear in an intermediate observation returned by the external environment. We implement the above variations of agent backdoor attacks on two typical agent tasks including web shopping and tool utilization. Extensive experiments show that LLM-based agents suffer severely from backdoor attacks and such backdoor vulnerability cannot be easily mitigated by current textual backdoor defense algorithms. This indicates an urgent need for further research on the development of targeted defenses against backdoor attacks on LLM-based agents. Warning: This paper may contain biased content.

cs.CR↗

arXiv2024

Nonlinear chiral metasurfaces based on structured van der Waals materials

Nonlinear chiral photonics explores nonlinear response of chiral structures, and it offers a pathway to novel optical functionalities not accessible through linear or achiral systems. Here we present the first application of nanostructured van der Waals materials to nonlinear chiral photonics. We demonstrate the three orders of magnitude enhancement of the third-harmonic generation from hBN metasurfaces driven by quasi-bound states in the continuum and accompanied by strong nonlinear circular dichroism at the resonances. This novel platform for chiral metaphotonics can be employed for achieving large circular dichroism combined with high-efficiency harmonic generation in a broad frequency range.

physics.optics↗

arXiv2024

On vector parking functions and q-analogue

In 2000, it was demonstrated that the set of $x$-parking functions of length $n$, where $x$=($a,b,...,b$) $\in \mathbbm{N}^n$, is equivalent to the set of rooted multicolored forests on [$n$]=\{1,...,$n$\}. In 2020, Yue Cai and Catherine H. Yan systematically investigated the properties of rational parking functions. Subsequently, a series of Context-free grammars possessing the requisite property were introduced by William Y.C. Chen and Harold R.L. Yang in 2021. %An Abelian-type identity is derived from a comparable methodology and grammatical framework. %Leveraging a comparable methodology and grammatical framework, an Abelian-type identity is derived herein. In this paper, I discuss generalized parking functions in terms of grammars. The primary result is to obtain the q-analogue about the number of '1's in certain vector parking functions with the assistance of grammars.

math.CO↗

arXiv2024

Towards Codable Watermarking for Injecting Multi-bits Information to LLMs

As large language models (LLMs) generate texts with increasing fluency and realism, there is a growing need to identify the source of texts to prevent the abuse of LLMs. Text watermarking techniques have proven reliable in distinguishing whether a text is generated by LLMs by injecting hidden patterns. However, we argue that existing LLM watermarking methods are encoding-inefficient and cannot flexibly meet the diverse information encoding needs (such as encoding model version, generation time, user id, etc.). In this work, we conduct the first systematic study on the topic of Codable Text Watermarking for LLMs (CTWL) that allows text watermarks to carry multi-bit customizable information. First of all, we study the taxonomy of LLM watermarking technologies and give a mathematical formulation for CTWL. Additionally, we provide a comprehensive evaluation system for CTWL: (1) watermarking success rate, (2) robustness against various corruptions, (3) coding rate of payload information, (4) encoding and decoding efficiency, (5) impacts on the quality of the generated text. To meet the requirements of these non-Pareto-improving metrics, we follow the most prominent vocabulary partition-based watermarking direction, and devise an advanced CTWL method named Balance-Marking. The core idea of our method is to use a proxy language model to split the vocabulary into probability-balanced parts, thereby effectively maintaining the quality of the watermarked text. Our code is available at https://github.com/lancopku/codable-watermarking-for-llm.

cs.CL↗

arXiv2024

Exploring Backdoor Vulnerabilities of Chat Models

Recent researches have shown that Large Language Models (LLMs) are susceptible to a security threat known as Backdoor Attack. The backdoored model will behave well in normal cases but exhibit malicious behaviours on inputs inserted with a specific backdoor trigger. Current backdoor studies on LLMs predominantly focus on instruction-tuned LLMs, while neglecting another realistic scenario where LLMs are fine-tuned on multi-turn conversational data to be chat models. Chat models are extensively adopted across various real-world scenarios, thus the security of chat models deserves increasing attention. Unfortunately, we point out that the flexible multi-turn interaction format instead increases the flexibility of trigger designs and amplifies the vulnerability of chat models to backdoor attacks. In this work, we reveal and achieve a novel backdoor attacking method on chat models by distributing multiple trigger scenarios across user inputs in different rounds, and making the backdoor be triggered only when all trigger scenarios have appeared in the historical conversations. Experimental results demonstrate that our method can achieve high attack success rates (e.g., over 90% ASR on Vicuna-7B) while successfully maintaining the normal capabilities of chat models on providing helpful responses to benign user requests. Also, the backdoor can not be easily removed by the downstream re-alignment, highlighting the importance of continued research and attention to the security concerns of chat models. Warning: This paper may contain toxic content.

cs.CR↗

arXiv2024

Room-temperature continuous-wave pumped exciton polariton condensation in a perovskite microcavity

Microcavity exciton polaritons (polaritons) as part-light part-matter quasiparticles, garner significant attention for non-equilibrium Bose-Einstein condensation at elevated temperatures. Recently, halide perovskites have emerged as promising room-temperature polaritonic platforms thanks to their large exciton binding energies and superior optical properties. However, currently, inducing room-temperature non-equilibrium polariton condensation in perovskite microcavities requires optical pulsed excitations with high excitation densities. Herein, we demonstrate continuous-wave optically pumped polariton condensation with an exceptionally low threshold of ~0.6 W cm-2 and a narrow linewidth of ~1 meV. Polariton condensation is unambiguously demonstrated by characterizing the nonlinear behavior and coherence properties. We also identify a microscopic mechanism involving the potential landscape in the perovskite microcavity, where numerous discretized energy levels arising from the hybridization of adjacent potential minima enhance the polariton relaxation, facilitating polariton condensate formation. Our findings lay the foundation for the next-generation energy-efficient polaritonic devices operating at room temperature.

physics.optics↗

arXiv2023

Two Stream Scene Understanding on Graph Embedding

The paper presents a novel two-stream network architecture for enhancing scene understanding in computer vision. This architecture utilizes a graph feature stream and an image feature stream, aiming to merge the strengths of both modalities for improved performance in image classification and scene graph generation tasks. The graph feature stream network comprises a segmentation structure, scene graph generation, and a graph representation module. The segmentation structure employs the UPSNet architecture with a backbone that can be a residual network, Vit, or Swin Transformer. The scene graph generation component focuses on extracting object labels and neighborhood relationships from the semantic map to create a scene graph. Graph Convolutional Networks (GCN), GraphSAGE, and Graph Attention Networks (GAT) are employed for graph representation, with an emphasis on capturing node features and their interconnections. The image feature stream network, on the other hand, focuses on image classification through the use of Vision Transformer and Swin Transformer models. The two streams are fused using various data fusion methods. This fusion is designed to leverage the complementary strengths of graph-based and image-based features.Experiments conducted on the ADE20K dataset demonstrate the effectiveness of the proposed two-stream network in improving image classification accuracy compared to conventional methods. This research provides a significant contribution to the field of computer vision, particularly in the areas of scene understanding and image classification, by effectively combining graph-based and image-based approaches.

cs.CV↗

arXiv2023

Communication Efficient Federated Learning for Multilingual Neural Machine Translation with Adapter

Federated Multilingual Neural Machine Translation (Fed-MNMT) has emerged as a promising paradigm for institutions with limited language resources. This approach allows multiple institutions to act as clients and train a unified model through model synchronization, rather than collecting sensitive data for centralized training. This significantly reduces the cost of corpus collection and preserves data privacy. However, as pre-trained language models (PLMs) continue to increase in size, the communication cost for transmitting parameters during synchronization has become a training speed bottleneck. In this paper, we propose a communication-efficient Fed-MNMT framework that addresses this issue by keeping PLMs frozen and only transferring lightweight adapter modules between clients. Since different language pairs exhibit substantial discrepancies in data distributions, adapter parameters of clients may conflict with each other. To tackle this, we explore various clustering strategies to group parameters for integration and mitigate the negative effects of conflicting parameters. Experimental results demonstrate that our framework reduces communication cost by over 98% while achieving similar or even better performance compared to competitive baselines. Further analysis reveals that clustering strategies effectively solve the problem of linguistic discrepancy and pruning adapter modules further improves communication efficiency.

cs.CL↗